Escaping the Valley of Disengagement : Two Field Experiments on Citizen Motivations to Engage in Collaborative Governance

Mark T. Buntaine UCSB

buntaine@bren.ucsb.edu

Jacob T. Skaggs UCSB

jacobtesoro@gmail.com

Daniel L. Nielson Brigham Young University dan_nielson@byu.edu

23 November 2016

Abstract

Governments face problems serving the public interest when they do not have good information about how well the demands of citizens are met. Citizens experience deficient or absent public services, but they do not have incentives to provide monitoring when they do not expect governments to be responsive to their concerns. Over time, this reinforcing cycle creates what we term the valley of disengagement. We investigate how to activate and sustain collaborative governance given the challenges posed by this vicious cycle. In two field experiments implemented in Kampala, Uganda, we recruited citizens to report on solid waste services to a municipal government. We find that community nominations of reporters and community announcements about reporters’ activity do not increase citizen monitoring. However, responsiveness to citizen reports by government significantly boosts engagement over several months, highlighting the critical role of timely and targeted responsiveness by governments for sustaining collaborative governance.

Acknowledgements

The authors are grateful to Polycarp Komakech, Immaculate Apio Ayado and Catherine Tabingwa for contributions to the design and implementation of this research. This project has been carried out in partnership with the Kampala Capital City Authority, and we gratefully acknowledge the support and participation of Charles Herbert, Josephine Kitaka, James Semuwemba, Martin Ssekajja, Frank Batungwa Tumusiime, and Judith Tukahirwa. Experiment 1 was supported by AidData at the College of William and Mary and the U.S. Agency for International Development (USAID) Global Development Lab through cooperative agreement AID­OAA­A­12­00096. The views expressed here do not necessarily reflect the views of AidData, USAID, or the United States Government. Experiment 2 was supported by the Hellman Family Foundation through a fellowship to MB. All activities described in this paper received approval from the University of California, Santa Barbara Human Subjects Committee (protocol ESMS­BU­MA­031), the Uganda Mildmay Research Ethics Committee (protocol 0706­2015), and the Uganda National Council for Science and Technology (protocol SS 3840) and the Uganda Office of the President (ref: ADM/154/212/03). We pre­registered the hypotheses and our plans for testing them at the Evidence in Governance and Politics registry ( 20151103AA ). The authors received helpful feedback on previous versions of this paper from Matt Potoski and from seminar participants at the University of California, Santa Barbara, and from conference participants at the 2nd Annual Conference on Environmental Politics and Governance in Gerzensee, Switzerland. The author contributions are as follows: MB is lead author. MB designed the research, with DN; JS implemented the research design, with MB; MB conducted the analysis; MB wrote the paper, with DN and JS.

1

Introduction

Governments — especially local councils and administrative offices — often lack good

information about where services should be extended, where existing public works are failing, and where

contractors or government employees are shirking. These information problems contribute to the

substandard provision of public services in both less­developed and developed countries around the

world. Citizens have information about absent or deficient public goods or services from direct

experience, but when they do not expect a response from government they do not have incentives to share

this information with government. Of course, governments around the world face variable incentives to

provide public goods to citizens contingent on the strength of democratic accountability, administrative

capacity, and oversight institutions. But even when governments are motivated to deliver public goods,

they cannot do so most effectively and efficiently without information from citizens.

We identify the key problem that surrounds initiating and sustaining collaborative governance in

which governments act on information provided by citizens about public goods and services. We term this

the valley of disengagement. When citizens do not expect government to be responsive to their concerns,

they have little incentive to engage in collaborative governance by providing monitoring or by supporting

the delivery of services. Lacking information and support, government cannot easily improve services

where they are in greatest demand, which over time reinforces distrust and disengagement. These

challenges limit the “long route” to accountability — actions by governments that respond to citizen

demands for better services and policies ( World Bank 2004 ).

Breaking out of the self­reinforcing pattern of disengagement and substandard service delivery

requires that citizens share the information they possess about public services in ways that lead to action

by government. We investigate how citizens can be motivated to provide information to government

about the quality of services in pursuit of better public goods. For example, where are water or electrical

outages most frequent? Did the mail arrive on time? What roads are in need of repair? Or, apropos of the

present study, did the providers of waste services pick up the trash this week? Governments could invest

in self­monitoring systems, but it is usually more efficient to rely on reports from residents, especially in

the age of widely available information and communication technologies. After all, citizens already

possess the necessary information. We thus extend to the mass populace McCubbins and Schwartz’s

seminal contribution about fire­alarm oversight ( 1984 ). Citizens can trip alarms that warn officials about

problems with public goods or services. The key is getting the citizens to share their information.

We theorize that citizens will share information they possess about public goods and services

when they have sufficiently positive beliefs about the responsiveness of government. We also hypothesize

that creating social benefits for reporting and selecting individuals that place greater value on attracting

2

public goods to their community will lower the level of belief about government responsiveness that is

necessary to activate and sustain citizen engagement in collaborative governance. We present a

decision­theoretic model that captures the beliefs that citizens hold over time about the responsiveness of

government and the social benefits that citizens obtain by engaging in collaborative governance as a

representative of a community. The model highlights the importance of providing sufficient opportunities

for citizens to positively update their beliefs about the responsiveness of government in order for

collaborative governance to take root. Without these opportunities, beliefs about the responsiveness of

government decline over time and engagement will cease once they fall below an activation threshold,

even when social benefits are available to citizens who value public goods. Once they become

deactivated, citizens do not have easy ways to positively update their beliefs about the responsiveness of

government, which entrenches disengagement and the substandard provision of public goods.

We test these theoretical predictions in two pre­registered field experiments conducted in close

partnership with the Kampala Capital City Authority (KCCA) related to the monitoring of solid waste

services. We prompted citizens of Kampala, Uganda to send reports over a number of months to the

KCCA about the management of solid waste services in their neighborhoods. Solid waste is a major

challenge in Kampala, with only a minority of waste produced in the city entering the formal waste

stream ( Kinobe et al. 2015 ). A large majority of Kampala residents are personally concerned with the

poor provision of waste services, as revealed in our baseline survey. The KCCA would like to improve

solid waste services but lacks efficient ways to collect information about the locations where service

delivery is substandard, information that citizens possess through their daily experiences. The main

outcome of interest in this study is the initial and sustained reporting of citizens about solid waste services

provided in their neighborhood. The outcome measure is the actual reports of citizens sent to the KCCA

from mobile phones, rather than self­reports about participation in collaborative governance as in related

research (e.g., Brabham 2009 ; Brabham 2012 ; Seidel et al. 2013 ).

We use three different treatment arms to test whether social benefits and the selection of reporters

who place great value on public goods increases reporting by citizens, which might offer a path out of the

valley of disengagement when citizens are already deactivated or might lower the threshold at which

citizens become deactivated — nomination of reporters by neighbors, nomination of reporters by

community leaders, and a community announcement about the work of reporters to improve solid waste

services by community leaders. Existing work highlights the importance of networks in creating

opportunities for engagement in policy­making and collaborative governance ( Berardo and Scholz 2010 ),

but we do not know if it is possible to leverage community networks to promote engagement by citizens

with government. To preview our results, even though we saw higher rates of reporting than any existing

3

study in Uganda that collected mobile feedback and indeed any other citizen reporting platform about

which we are aware (e.g., Blaschke et al. 2013 ; Grossman et al. 2014 ), with approximately 20% of

reporters sending reports during our study period, we did not find evidence that any of the recruitment or

announcement conditions increased short­term or long­term engagement in collaborative governance by

citizen reporters. This is good news from a policy perspective, because costly recruitment and social

motivation treatments appear to be unnecessary to produce more engagement by citizens.

We also experimentally varied responsiveness by the KCCA to citizen reports to test our

prediction that rapid, timely, and targeted responsiveness is key to activating and sustaining citizen

reporting. Reporters from neighborhoods in the responsiveness treatment received a weekly, targeted

announcement about how reports are being translated into official action plans and used to improve solid

waste collection in Kampala. We know of no other research that experimentally varies responsiveness to

citizen concerns by government, despite the core role that trust and the building of collaborative ties

between agencies and citizens play in theories about collaborative governance ( Ansell and Gash 2008 ;

Sandström et al. 2014 ). We find that government responsiveness significantly boosts long­term

engagement of reporters over months, as measured by actionable and usable reports. This result highlights

the importance of continuously supporting citizens’ beliefs about responsiveness. Building responsiveness

into governance arrangements across a variety of settings might significantly increase participation by

citizens in improving the delivery of public goods. An endline survey that we fielded five weeks after the

reporting period did not reveal increased trust in government or satisfaction with services among reporters

in the responsiveness condition, indicating the importance of continuously reinforcing responsiveness

when attempting to motivate citizens to engage in the monitoring of public services.

The results of these field experiments are particularly significant against the backdrop of an

expanding set of information and communication technologies (ICT) that raise the possibility for

low­cost, targeted, and timely responsiveness by governments, enabling a virtuous cycle of collaborative

governance with citizens. In places where the capacity of governments is low and the management of

public services is poor, building the responsive relationships with citizens necessary to promote

collaborative governance is especially difficult. Citizen do not engage because they perceive governments

to be unresponsive to their concerns. Governments cannot improve public services because they then do

not have information on the demands of citizens or access to monitoring. Finding ways to engage citizens

in collaborative governance is vitally important across a wide range of functions that fall to governments.

Indeed, community policing depends on building trust between citizens and the police and facilitating the

flow of information necessary to act on crime ( Brogden and Nijhar 2005 ). Education is enhanced by

involving parents in the local management of schools and monitoring of teachers ( Duflo et al. 2015 ).

4

Water user boards augment the collaborative management of collective water resources ( Berkes 2009 ).

There is great optimism that ICT can improve collaborative governance in these areas and others

by increasing citizen participation, generating more useful information about public demands for services,

and allowing governments to more effectively and rapidly respond to the concerns of citizens ( Linders

2013 ). Governments around the world are building platforms to collect information from citizens to

improve the provision of public services ( Smith and Reilley 2013 ). They enlist citizens in collaborative

governance when it comes to policing, the maintenance of infrastructure, and planning for services. Some

evidence suggests that ICT can broaden public participation in governance ( Grossman et al. 2014 ). Yet

other efforts have failed to activate and sustain citizen engagement ( Evans & Campos 2013 ; McGee and

Carlitz 2013 ; Grossman et al. 2015 ; Grossman et al. 2016 ). By harnessing the strengths of new modes of

citizen­government interaction to foster information sharing and responsiveness, it is likely possible to

escape the valley of disengagement in a variety of settings. But building these new ICT platforms is not

enough to sustain collaborative governance; citizens need to understand how their effort is rewarded in

terms of responses from public officials.

Our theory and results break ground in understanding how collaborative links between citizens

and governments that contribute to good governance are built. While the policy influence of voters on the

choices of elected representatives ( Przeworski and Stokes 1999 , Powell 2004 ) and the role that

information plays in promoting electoral accountability ( Besley and Burgess 2002 ; Adserà, Boix and

Payne 2003 ; Ferraz and Finan 2011 ) have received much attention in research about politics and

governance, the collaborative links between citizens and government offices apart from elections may be

similarly fundamental for governance and arguably more important for the provision of public goods and

services. Our results offer strong evidence about the dynamics of building ongoing, collaborative

relationships between citizens and governments to provide public goods.

Background and Theory

Responsive governance requires that citizens learn about the outcomes of actions by government

and possess the collective ability to remove politicians for poor performance or to reward good outcomes

with longer tenure. But the information that voters hold about politician performance when they go to the

polls is not sufficient for governance that is responsive to the demands of citizens. Information must flow

from citizens to governments in greater detail than can be conveyed by election returns. Before the many

goods and services that citizens demand from government are actually delivered, long chains of

delegation between citizens, politicians, bureaucrats, and contractors organized into multiple layers must

be formed and moved to action. Agent slack and slippage are likely at every link in the chain, and so

5

extensive information must be gathered on the behavior and resulting outcomes of agents at each stage

( Nielson and Tierney 2003 , Hawkins et al. 2006 ). Monitoring and oversight from the top down can prove

very expensive ( Kiewiet and McCubbins 1991 ), so bottom­up information is often sought as a low­cost

and efficient solution to learning how frontline government units and their contractors are performing

( McCubbins and Schwartz 1984 ). The information requirements for good governance are especially

intense at the local level, where thousands of local governments provide goods and services to millions of

people. Citizens possess critical information about the status of public goods and services, but they need

to be motivated to pay the costs of providing the information to the government office that has the

primary authority for providing the public good in question.

If citizens do not believe that there will be a response to the information that they could provide,

then they will not have incentives to engage in monitoring as part of collaborative governance. Indeed, the

record of transparency and accountability initiatives that involve citizen monitoring of governments is

mixed (for recent reviews, see Joshi 2013 ; Fox 2015 ), which means that in many settings citizens are

correct in believing that government will not be responsive to information that they could provide. Insofar

as there is any consensus about the reasons why initiatives to solicit information and participation from

citizens do not always improve service delivery, it is that citizens lack ways to motivate governments or

services providers to act on the information that they provide ( Banerjee and Duflo 2006 , 124; Banerjee et

al. 2004 ; Olken 2007 ). In contrast, some studies about citizen monitoring in settings where rewards and

punishment mechanisms are available to citizen monitors — whether through social pressure on providers

( Bjorkman and Svensson 2009 ) or by enhancing enforcement by government ( Caseley 2006 ) — have

found citizen monitoring to be effective at improving public services.

In a recent reflection on this body of mixed evidence, Fox (2015) offers the important critique

that many studies of citizen monitoring and government accountability are “tactical” rather than

“strategic” — that is, they consider mostly how to get information flowing in one direction, rather than

strengthening feedback loops between citizen monitors and governments, which might sustain

engagement in collaborative governance and social accountability. As Mansuri and Rao ( 2013 ) highlight

in another review, the longer term success of social accountability and collaborative governance schemes

depends on governments’ using their abilities in sanctioning and oversight to act on the information

gained through citizen engagement. However, considering the strategic context, where public services are

substandard and, as a consequence, trust in government is low, the prior beliefs of most citizens likely

hold that government will not be responsive to their concerns, which will tend to promote lower quality

monitoring and to decrease engagement by citizens.

6

A decision­theoretic model of reporter behavior

Consider the simple illustration of a government agency and k citizens C 1 , C 2 , …, C k who might

engage in the collaborative governance by reporting on the status of services. In this setting, each citizen

has uncertain beliefs about whether government is responsive or nonresponsive to reports that they could

potentially submit. We denote the true probability that government will respond to a report by . Each

citizen makes a decision about whether to report on public services ( R kt ) as a function of their belief about

the responsiveness of government at a given point in time p kt (), the value of a government response to

the individual v k (which might include prosocial utility), the cost of the reporting c k , and any outside

social benefits b k such as prestige among community members or the positive feeling of status gained

being responsible for the well­being of the neighborhood. Each citizen will report at time t if:

(1)

Before proceeding to simulate how this belief and incentive structure drive the dynamics of collaborative

governance, we consider each of the component parts and their place in existing theory:

Responsiveness from government (p kt () )

Even if government is a responsive type, it can be very difficult for citizens to observe and

attribute responsiveness to government. If citizens already have low expectations of governments, it is

likely that the prior beliefs of citizens about the likelihood of a governmental responses to their concerns

will likewise be low. These beliefs will result in disengagement, which will provide few opportunities for

citizens to update their beliefs through direct experience, even if government is the responsive type. Other

reviews of participatory governance highlight how costs to citizens can undermine public participation in

governance (e.g., Speer 2012 ), but the role of beliefs about the responsiveness of government and the

ways these beliefs are likely to change when citizens interact with governments have not been theorized.

Using survey data, Tolbert and Mossberger show that citizens who interact online with governments

generally have higher trust in government, perhaps through “interaction[s] with officials that are

convenient and quick, potentially enhancing responsiveness” ( 2006 ). They show an increased satisfaction

in interactions with government when citizens visit government websites, but the reasons for this effect

are not clear and the methods that governments might employ to get citizens to engage in collaborative

governance, especially when engagement goes beyond the consumption of information, are not explored.

7

In settings where institutional arrangements and political conditions make governments responsive, the

core challenge of fostering collaborative governance, we argue, is cultivating citizen beliefs about

responsiveness that match the actual level of responsiveness from government.

Closer to the setting of our own study, Grossman et al. (2016 ) ask how citizens can be motivated

to report deficiencies in public services to local politicians in Uganda. They argue that one of the primary

challenges of promoting collaborative governance using communication technologies is overcoming low

levels of efficacy among citizens who experience deficiencies. They theorize that most citizens in Uganda

and other developing countries have low “external efficacy,” defined as a lack of expectations about “the

responsiveness of government authorities to citizens’ demands writ large” (3). As an experimental

treatment, they send messages to subjects from local officials encouraging the reporting of deficient

public services and find that the rate of ever­participation — citizens that use the platform at least once

over a six­month period — rises from approximately 3.4 percent in control to 4.7 percent in treatment.

We take this idea further by manipulating not just encouragement from politicians to report on

public services, but instead informing citizen reporters exactly what the governmental agency receiving

their reports is doing in response on a weekly basis. Our responsiveness treatment included making

weekly action plans for the mobilization and oversight of contractors in specific zones, organizing

systematic zone­wide clean­up, and engaging in new public outreach campaigns. It has been theorized

that the key to activating and sustaining collaborative governance is to build trust between citizens and

government ( Ansell and Gash 2008 ). Our treatment is designed to directly increase citizen beliefs about

government responsiveness and perhaps increase the “external efficacy” of citizens in engaging in

collaborative governance. By varying not just encouragement, but actual responsiveness from a

government agency to concerns, we more directly address key theoretical predictors of longer­term

engagement of citizens in collaborative governance.

Social benefits gained by reporting (b)

Previous research has found that the social benefits gained through networks are often critical to

long­term engagement in social monitoring, especially when participation takes place through the internet

or other digital communication tools. Perceptions of the individual or collective benefits gained through

engagement ( Budhathoki and Haythornthwaite 2013 ; Chandler and Kapelner 2013 ) and opportunities for

social connections ( Brabham 2009 ; Blaschke et al. 2013 ) drive the digital engagement of citizens in

public life. Accordingly, we investigate whether nominations by neighbors or community leaders or

announcements about the work of reporters by community leaders, and the social expectations and status

that are likely to come with these actions, drive engagement by citizens in collaborative governance.

8

Research on the role of networks in collaborative governance finds that networks create opportunities for

engagement by citizens and civil society ( Berardo and Scholz 2010 ), but less is known about leveraging

pre­existing social and community networks to initiate and sustain the pursuit of public goods in

collaboration with governments. Initial evidence from experiments in Uganda suggest that social benefits

outside of the specific public goods in question can significantly elevate participation in SMS­based

platforms that promote discussion of public issues ( Blaschke et al. 2013 ). In the present setting,

nomination may make nominees feel a greater social obligation to accurately represent public services

and identify problems on behalf of their community, since a fellow member or leader of their community

has raised them to this position of status.

Value of government response to the individual (v)

Another way to increase the engagement of citizens in collaborative governance is to find citizens

who value attracting public goods provided by government to their community. We thus also expect

nomination by neighbors and community leaders can select for individuals who place higher value on

improving public services. Because individuals who make nominations can maximize public goods by

nominating reporters with pro­social and leadership attributes, we expect that nomination will enhance the

provision of citizen monitoring. Existing research on referrals and nominations have mainly been studied

in labor markets, rather than situations where individuals are asked to produce information that is useful

for governance. Studies of referrals in labor markets have found that referred employees exhibit lower

quality than non­referred employees, potentially due to an incentive to trade favors within social networks

( Fafchamps et al. 2015 ; Beaman and Magruder 2012 ). These results have not been replicated for citizen

reporting on public services, where nominations are made to gain public rather than private goods. While

our experimental design did not allow us to parse out how nomination influences separately social

benefits and the selection of individuals who place a higher value in attracting public goods to the

community, we expected that nomination would enhance reporting via both mechanisms.

While not the theoretical focus of the present investigation, the salience of routine public services

is also an open question. Citizens divide their attention across many diverse issues. Many existing

instances where citizens send feedback to government have come about after disasters ( Crooks & Wise

2013 ). In these contexts, it is easier to gain enough dispersed attention from citizens so that collective

information is provided. It may be the case that routine public services do not generate high salience.

Indeed, most platforms designed by organizations and agencies to collect external information fail to

generate sustained engagement and these failures often do not appear in the published literature ( McGee

and Carlitz 2013 ; Dahlander and Piezunka 2014 ). In the present case, we chose to focus on a public

9

service where we have primary evidence about the concern of citizens about its provision such that we are

unlikely to see very low rates of reporting. Future research will need to better deal with the role that issue

salience plays in the mobilization of citizens to participate in collaborative governance.

Cost of reporting (c)

The core source of optimism about collaborative governance in the information age is an

opportunity structure that is more open to citizens. Across a variety of places and settings, “opportunity

structure” influences citizen participation in social movements, policy­making, and collaborative

governance ( Stevenson and Greenberg 2000 ; Leifeld and Schneider 2012 ; Vráblíková 2014 ). Features of

each setting, such as linkages in political networks, the costs of communications, and the rules of

decision­making, influence decisions by citizens about participation in the pursuit of shaping governance.

The idea that information technologies can facilitate citizen engagement in the governance of public

services has sparked enthusiasm because it alters the opportunity structure, primarily by substantially

decreasing the costs of sharing and processing information relevant to public life ( Oates 2003 ; Grossman

et al. 2014 ; McGuire 2006 ; Charalabdis et al. 2012 ; Linders 2012 ; Rotberg & Aker 2013 ). Unlike

collaborative governance via traditional means, which often involves significant time and costs for

citizens, mobile phones allow for instantaneous and, in many deployments, toll­free access to public

officials. Of course, there are still barriers to engagement, including literacy, monitoring costs, knowledge

about how to use communication technologies, and time, but these barriers are also likely to affect

traditional means of engagement. Our study operates within this context, but extends and contextualizes

what is required to take advantage of the changing opportunity structure provided by ICT.

Simulation of the reporter beliefs and reporting over time

Considering the model above and its component parts, it is straightforward that adding social

benefits to the reporting process or decreasing the cost of reporting will encourage more engagement in

collaborative governance. What is missing and more interesting is the time path of p kt () as each reporter

perceives responsiveness or a lack thereof to their engagement. Consider a standard Bayesian updating

model where the prior p kt () takes the form of a beta distribution, where p () = ­1 (1­) ­1 . In each

period t , if and only if a reporter submits a report, then they will have the opportunity to perceive with

error whether government responds to the report. More formally, after submitting a report, the citizen

views the outcome of a Bernoulli trial screened by an error function (), which might not have an

expected value of , such as when any response is not easy to attribute to the government or if the

government is able gain positive credit even though it is not responsive. We now have the machinery to

10

simulate the reporting behavior of citizens over time.

Consider a set of reporters who have various prior beliefs about the responsiveness of

government, when the true value of indicates that government is in fact the responsive type. In the case 1

of a government that has difficulty quickly enacting a response to public reports, we use the simple

adjustment where () is the outcome of the Bernoulli trial with probability , filtered by some delay in

the opportunity to observe responsiveness to reports. Of course, () may take on a variety of functional

forms that we do not attempt to fully anticipate here, but a delayed or unobserved response is perhaps the

most typical type of error process that citizens experience when they interact with government. Recall that

once the value of p kt () drops below a certain level, determined by the other parameters in the decision

model, the reporter will not submit reports and will therefore not have the chance to update their beliefs

about responsiveness. If social benefits are added, such as prestige among the community, the value of

p kt () can become lower before a reporter will become deactivated. Figure 1 shows simulated paths of the

mean value of p kt () at each period, which is equal to the posterior value of the previous period after

taking into account the result of (), which in this case delays the observation of any response. We set

other parameters in (1) so that reporting ceases when p kt () falls below an activation threshold that is

significantly lower than the actual responsiveness of government . In the left column of Figure 1, no

social benefits are added, which raises the value of p kt () needed to sustain reporting as compared to the

right column, where social benefits lower the value of p kt () needed to sustain reporting.

1 In the simulations below, we set = 0.8, the reporting activation threshold based on the other parameters to 0.4, assume the value of b k is 0.1 for all reporters k , and draw the starting beliefs of reporters p k,t=0 () randomly from a uniform distribution of mean values [0.1,0.9].

11

Figure 1. The posterior beliefs about the responsiveness of government when observations about responses to citizen reports are delayed by the number of periods indicated. Red lines indicate reporters who do not fall below the activation threshold during the simulation period and grey lines indicate reporters who do fall below the activation threshold.

As displayed in Figure 1, any delay in observing the response of government to citizen reports

decreases posterior beliefs that government is the responsive type. This means that only citizens with very

high prior beliefs about the responsiveness of government will persist in reporting (red simulated paths),

12

while all others will fall below the activation threshold and fail to update further (grey simulated paths).

When social benefits are available to reporters (Figure 1, right column), the threshold at which reporters

become deactivated due to their beliefs about government responsiveness is lower, which also promotes

more long­term reporting. These dynamics illustrate the critical role that targeted responsiveness is likely

to play in keeping beliefs about responsiveness high enough among at least some citizens that their voices

matter and that they should continue participating in collaborative governance.

Pre­Registered Hypotheses

Based on the expectation that nomination can enhance the provision of reporting about solid

waste services either by raising social benefits (b k , H1­H3) or by enhancing beliefs about the

responsiveness of government ( p kt (), H4), we pre­registered the following hypotheses prior to randomly

assigning recruitment conditions or collecting any data on outcomes (SI, Appendix D contains the exact

wording of pre­registered hypotheses, which are shortened here for readability):

H1: Nomination by neighbors will increase reporting.

H2: Nomination by the local council chair will increase reporting.

H3: Announcement by the local council chair about the reporters will increase reporting.

H4: Responsiveness to citizen reports will increase reporting.

Experimental Design

We designed and carried out two randomized field experiments to understand whether social

motivations and government responsiveness can initiate and sustain participation by citizens in the

collaborative governance of public services. In particular, the treatments that we employ are meant to

raise either beliefs about responsiveness ( p kt ()) or the social benefits (b k ) of engaging in collaborative

governance. We focus on citizen reporting about solid waste management, which generates high levels of

citizen concerns, with 90% of residents in our study area personally concerned with the state of solid

waste management as of 2014 (see SI, Appendix A for results of a pre­experimental baseline survey

undertaken to scope out this project).

13

Setting

Kampala, Uganda faces similar problems of monitoring and accountability for solid waste

management as many other parts of the world ( Bhuiyan 2010 ; Okot­Okumu and Nyenje 2011 ). With

Kampala growing rapidly like many developing cities ( Vermeiren et al. 2012 ), the need to improve the

quality and scale of services is pressing. Private companies contracted to remove solid waste often

provide services of lower quality to groups of people that are not able to share monitoring information

( Oteng­Ababio et al. 2010 ; Katusiimeh et al. 2012 ). Since most of Kampala is contracted to private

collectors, city managers find themselves in a challenging position, especially given information

asymmetries, pressures toward corruption, and wealth disparities across communities.

Our close partner in this project, the Kampala Capital City Authority (KCCA), has prioritized

improving solid waste management in order to boost resident satisfaction and promote public health.

Kampala is also one of the key strongholds of opposition support in Uganda, and the nationalized KCCA

has a strong political mandate from the ruling party to improve resident satisfaction with government

services. Additionally, the KCCA has been supported by international donors for over a decade to

improve waste management, but still finds it difficult to engage the public in actionable ways. Despite

having used public resources to develop an interactive SMS platform and a KCCA mobile application to

exchange information with citizens, the KCCA struggles to use its technological investments to exchange

useful information with the public. They now seek to understand whether mobile technologies can

enhance public engagement and encourage more accountable provision of public services.

Phase 1 Experimental Design

In Phase 1, we recruited 1040 citizen reporters in 90 administrative zones to provide feedback on

solid waste removal services and disposal practices at the spatial scale of neighborhoods. In November

2015, our team of enumerators carried out a recruitment drive over a period of two weeks to form our

experimental sample of reporters. The KCCA provided us with a list of all zones (LC I) inside the capital

city jurisdiction of Kampala and the associated shapefiles outlining their boundaries. At the time of the

first experiment, there were a total of 755 zones (LC I) contained within 97 parishes (LC III) and 5

divisions used to manage waste services. We randomly selected 90 zones for our experimental sample. We

dropped 11 zones from the original random sample because they were demolished at the time of

enumeration, the enumeration team was not able to locate any residences within the zone, or the

enumeration team was not able to gain access to gated zones within the diplomatic district. We replaced

these 11 zones with another random sample to form the final experimental sample.

After selecting the experimental sample, we randomly assigned each zone to one of two reporter

14

recruitment conditions. In each zone, we then aimed to recruit 12 citizen reporters according to the

recruitment condition assigned at the zone level, for a total of 1040 reporters (see SI, Appendix B for

detailed recruitment protocols):

(Recruitment Condition 1) Random Citizen recruitment : Following a random walk pattern, the

enumeration team approached adults walking or sitting outside of their homes or businesses and

asked whether they would sign up to be a reporter.

(Recruitment Condition 2) Neighbor Nomination recruitment. Following a random walk pattern,

the enumeration team approached adults walking or sitting outside of their homes or businesses

and asked whether they could nominate a “trustworthy and responsible” individual who lives in

the zone to report on behalf of its residents. If the individual indicated willingness to make a

nomination, the enumerator asked the citizen to make a face­to­face introduction to the nominated

individual. This nominated individual was then asked whether they would sign up to be a reporter.

Over a 7­week period following the recruitment drive, all citizen reporters received prompts from

the KCCA’s interactive SMS messaging system in the same way. Reporters received a total of 17 prompts

for information about waste pick­up schedules, waste burning practices in their zone, and the locations of

waste piles that needed special attention by the KCCA or its contractors (See Appendix C for a list of

prompts). To minimize the costs of reporting, we sent an airtime credit initially to all reporters and then

also sent credit each week to the phones of all subjects who submitted at least one response to the prompts

that week. To further encourage reporting, we held a lottery for a ~$10 prize in airtime each week for all

reporters who participated. Further details and justification for the implementation procedures are

contained in our publicly­available pre­analysis plan ( EGAP registered design 20151103AA ).

Phase 2 Experimental Design

In June 2016, our team of enumerators recruited an additional 1,905 reporters from 97 randomly

selected administrative zones or local councils (LC 1) within the KCCA’s waste management service

zones. Ninety­six zones were included in the original experimental sample of zones. Five of the original

zones were dropped becauses they could not be located or were demolished. Six randomly selected zones

were added to replace the zones that were either dropped or had low recruitment numbers due to the small

size of the zone. In each zone, we aimed to recruit 20 reporters. Each zone was divided into four cells of

roughly similar geographic size and five individuals were recruited to be reporters from each cell.

15

Reporters were required to be adult residents of the zone and be the primary user of their own mobile

phone. The zones did not overlap with the Phase 1 sample (Figure 2).

With the zone as the unit of randomization, each zone was assigned one of eight different

treatment combinations based on a three ­factor crossed experimental design. Two factors were

recruitment and announcement conditions, while the third factor was a condition that highlighted the

responsiveness of government to citizen reports, as follows:

(Recruitment Condition) Random Citizen recruitment : Following a random walk pattern, the

enumeration team approached adults walking or sitting outside of their homes or businesses and

asked whether they would sign up to be a reporter. This condition follows exactly the protocol

from Phase 1.

(Recruitment Condition) LC1 Nomination recruitment. Reporters in zones assigned to the LC1

nomination condition were recruited by the local council chairperson (LC1) or a delegated

zone ­level authority figure. We chose LC1s to select citizen monitors because they provide

oversight and hold authority at the zone level, which may heighten the desirability of providing

frequent and high ­quality information about waste services for reporters. Further, LC1s are

typically well known throughout their zones and likely know the community members well

enough to select citizen monitors that will represent their zone well. LC1s nominated reporters by

introducing them to the recruitment team.

(Announcement Treatment) Announcement of Reporters by LC1. Reporters in zones assigned to

this treatment condition were informed that the LC1 would announce the citizen monitoring

program and the names of reporters at an upcoming zone ­wide meeting. After all 20 reporters

were recruited in a zone, a list of the names of selected citizen monitors and information on the

program were left with the LC1. The implementation team contacted LC1s by phone one week

following the completion of the recruitment activity to remind the LC1s to make the

announcement at a community meeting. If the LC1 was unavailable during the recruitment drive, 2

the recruitment responsibilities were delegated to another local authority figure such as the

2 We collected data on compliance with the announcement treatment and found that only 38% of the community leaders in zones who were assigned to this condition and who we were able to contact at endline delivered the announcement treatment. The reporters in these zones still expected a community announcement, since they were fully informed about the upcoming announcement during recruitment, so our main analysis below is intent­to­treat. In SI Appendix G, we estimate complier average causal effects for the announcement treatment by 2SLS.

16

Vice ­Chairperson, Secretary of Defence, or member of the Local Council Committee. If the

recruitment responsibility was delegated, the LC1 was contacted that day to inform him/her about

the program, the recruitment task and to whom it was delegated, and any other additional

responsibilities they had relating to the citizen monitoring program such as the announcement

treatment. Zones not assigned to treatment were assigned to a control condition where the LC1

was not requested to make an announcement.

(Responsiveness Treatment) Responsiveness from the KCCA. In a third experimental condition,

reporters received weekly personalized text messages informing them that their responses to

prompts had been sent to the KCCA Waste Management team, and communicating the real action

plans that had been made by the KCCA Waste Management team on the basis of reports.

Reporters were also sent information listing the number of responses they sent that week and the

total number of responses sent by all citizen monitors in the reporter’s zone that week, and

offering to answer any questions about how the data was being used by the KCCA. Any 3

questions were answered during a call center held each week. During phone calls with subjects in

neighborhoods assigned to this treatment, our team reemphasized the objectives and purpose of

the citizen monitoring platform outlined in the reporter sign­up form. Also as part of the

responsiveness treatment, subjects were contacted through voice calls on their mobile phones one

month following the start of the reporting period. The program representative discussed the

quantity of the subject’s responses, reminded them of objectives and expected results of the

program, and explained how the subject’s reports are being used to improve waste management in

their zones based on action plans developed by the KCCA in response to citizen reports. Both

active and inactive reporters received the responsiveness outreach. Zones not assigned to

treatment were assigned to control and the reporters did not receive any messages or phone calls

about what the KCCA was doing with their reports. For zones in the control condition, the KCCA

3 The intention of telling subjects how many reports were received from the specific individual and the whole zone during the previous week was heighten the sense that the reports were being noticed and used. One concern about this treatment design, however, is that it may introduce a bundled social norms treatment into the Responsiveness treatment. We are not very concerned about this possibility because reporters did not know the number of other reporters in the zone and could therefore not easily interpret the number of responses as a social norm. Regardless, past research has found that subjects tend to revert to the social norm when they are provided information about a social norm ( Schultz et al. 2007 ). In our case, if the social norm was exerting an effect we should see reporters who were active the previous week become less active in the subsequent week. In SI Appendix H, we show instead that the Responsiveness treatment continues to exert a positive treatment effect considering only the subset of reporters who were active during the previous week, which rules out this concern.

17

asked only to receive a digest of reports at the end of the reporting period and did not respond

weekly to reports.

Figure 2. Final sample of zones within the jurisdiction of the Kampala Capital City Authority for Phases 1 and 2 recruitment drives and the combined responsiveness treatment condition. This is the final random sample, after replacing zones that did not contain residences or that were inaccessible.

During the 8­week Phase 2 reporting period between July 2 and August 29, 2016, all subjects

recruited during Phase 1 and Phase 2 were sent 15 prompts for reports. The questions we asked our

subjects were based on information that the KCCA identified as most useful in monitoring the quality of

services provided by its waste management contractor. Prompts included general questions about

zone­level waste conditions and the quality, frequency, proximity of waste collection services provided to

the zone, and several open­ended questions (See Appendix C for the list of prompts used in Phase 2). As

in Phase 1, we encouraged reporters to answer prompts by running a lottery each week for ~$10 in

airtime. Three weeks after the end of the Phase 2 reporting period, we implemented a short survey to

understand whether responsiveness increased trust in government and satisfaction with waste services,

which would indicate longer­lasting shifts in more general attitudes as a result of responsiveness.

Summary of experimental design and conditions

There was a break of almost six months between the two phases when no prompts were sent to

reporters from Phase 1. Figure 3 summarizes the combined design of the two experiments reported here.

18

Figure 3. Diagram of experimental design for both Phases 1 and 2

Outcome Measures of Reporting

As pre­registered, we measure increased reporting in a number of different ways: (1) The total

number of active reporters (i.e., those submitting at least one report) during the reporting period; (2) The

total number of reports submitted by each reporter during the reporting period; (3) The total number of

reports submitted by each reporter during the last two weeks of the reporting period; and (4) The total

number of open­ended reports (e.g. descriptions of location of piles) submitted by each reporter during the

reporting period.

Descriptive Data on Reporters

The reporters in our study are likely to be fairly representative of Kampala residents, since many

of the recruitment conditions began with random walks in randomly selected zones around the city. It may

be the case that the nomination process produced reporters of a different type on observable

characteristics, but we do not find strong evidence for this possibility (Table 1). The only notable

exception is that LC1 nomination produced reporters with longer average periods of residence in the zone

than did any of the other recruitment conditions. In order to avoid Hawthorne effects, the reporters were

asked only to provide brief information for intake into the KCCA reporting system, rather than a full

survey of demographic and attitudinal responses that would have required a different informed consent

19

process for research subjects. All reporters were fully informed that the platform was being operated and

tested with the KCCA.

Table 1. Characteristics of reporters in both Phases

Phase 1 Random Recruitment Neighbor Nomination

Years in zone (mean) 9.24 9.15

Female (proportion) 0.39 0.45

Age (mean) 30.2 30.8

Satisfied with waste services (proportion) 0.28 0.32

Phase 2 Random Recruitment LC1 Nomination

Years in zone (mean) 11.0 15.2

Female (proportion) 0.62 0.65

Age (mean) 32.4 36.0

Satisfied with waste services (proportion) 0.36 0.36

Analytical Methods

As we pre­registered, we performed hypothesis tests via randomization inference for difference in

means between experimental conditions. We assume the sharp null hypothesis (no unit­level treatment

effects) such that Y i (1) = Y i (0) for all zones or reporters where Y i (1) is the potential outcome if assigned

to nomination and Y i (0) is the potential outcome if assigned to random recruitment. We then generate

5,000 iterations of our exact clustered randomization procedure and capture the sampling distribution of

treatment effects observed under the sharp null. We compare the observed difference in the value of

interest between treatment conditions and compare that value to the sampling distribution to compute a

p ­value of how often such a difference would be observed by random chance. For the Phase 2 analysis,

because of the ease of reporting on multiple treatment arms, we estimate the effects of treatment at the

reporter­level via OLS regression. We have confirmed that the substantive and statistical significance of

all effects are robust to the pre­registered difference­in­means specifications. Appendix F contains the

same Phase 2 results with analysis performed at the zone level. We also observed significant

non­compliance with the LC1 Announcement treatment in Phase 2, prompting us to estimate complier

average causal effects as a robustness check on the intent­to­treat results reported below (see SI Appendix

G). In no case does this change the substantive or statistical significance of the main results.

20

Findings: Phase 1

In the first experiment, we did not find any significant difference in the amount of reporting by

citizen reporters recruited randomly or by neighbor nomination, that latter procedure being intended to

raise the social benefits of participating in collaborative governance for reporters (Figure 4). In total, we

received 528 SMS reports that were on­topic and contained information relevant to solid waste

management. In both the Random Citizen and Neighbor Nomination recruitment conditions, we observed

similar proportions of reporters who submitted at least one report during the 7­week reporting period, at

approximately 15% (Panel A; te =0.010, p=0.34). If we instead compare the mean number of responses

per reporter by recruitment condition, we find that nominated reporters submitted an average of 0.490

reports, while randomly recruited reporters submitted an average of 0.460 reports, which is consistent

with variation expected under the null hypothesis (Panel B; te = 0.030, p=0.38). Finally, if we consider

how many times reporters responded to open­ended prompts for the locations of trash piles, potentially

the most costly type of reporting in terms of effort, we again find almost equal levels of reporting across

recruitment conditions (Panel C; te =0.010, p=0.29).

Figure 4. Reporting by recruitment condition during Phase 1. (A) Proportion of reporters who submitted at least one report by recruitment condition; (B) Average number of total reports per reporter by recruitment condition; (C) Average number of open­end reports per reporter on the location of waste piles by recruitment condition. No significant differences in reporting between recruitment conditions identified. All panels display one standard error bars.

21

Findings: Phase 2

In the second experiment, we examine the same three outcomes as a function of the three

recruitment and treatment conditions. Recruitment and announcements by LC1 chairs were intended to

raise the social benefits of reporting, while the responsiveness treatment was intended to influence beliefs

about the responsiveness of government. We report results both for the pooled group of subjects recruited

during both Phase 1 and 2, as well as the results split by the recruitment phase. Considering first the

number of reporters during Phase 2 who submitted at least one, on­topic report about solid waste

management during the eight­week period, only the Responsiveness condition boosts participation (Table

2). Reporters recruited during Phase 1 from a zone assigned to the responsiveness condition are 50% more

likely to be active during Phase 2 than reporters in control zones. Reporters recruited during Phase 2 from

a zone assigned to the responsiveness condition are 14% more likely to be active than reporters in control

zones. This result indicates that hearing about what the government is doing with the reports via targeted

outreach can help initiate engagement in citizen reporting. In contrast, we do not observe any differences

in the number of active reporters when recruiting by either neighbor or LC1 nomination, or when

reporters expected the LC1 to make an announcement about the platform and reporters’ names at a

community meeting. Thus, the evidence suggests that social motivations are not effective at activating

reporting on public services in this context, which is good news for policymakers who do not need to

spend extra resources on recruitment and social motivation to promote collaborative governance.

22

Table 2. Total number of active reporters during Phase 2

Turning to the total number of reports made by each reporter during the 8­week Phase 2 reporting

period, we find very similar results, with only the responsiveness treatment driving more reports (Table

3). Pooling zones across recruiting periods, we find that the Responsiveness treatment increased the

average number of reports per reporter by approximately 0.4 over eight weeks. This result is largely

driven by the significant effect that the Responsiveness treatment had on treated Phase 1 reporters, among

whom the Responsiveness treatment increased the number of total reports per reporter by 83%. In

contrast, the Responsiveness treatment did not increase the total number of reports by Phase 2 reporters in

ways that are highly inconsistent with random chance (for P2 Reporters model, p=0.12). Like the results

for active reporters, we do not observe any differences in the number of reports per reporter when

recruiting was done by either neighbor or LC1 nomination, or when reporters expected the LC1 to make

an announcement about the platform and reporters’ names at a community meeting.

23

Table 3. Total number of reports submitted by each reporter during Phase 2

Finally, we consider the total number of reports by each reporter the last two weeks of the 8­week

reporting period. As pre­registered, we are interested not only in the total effects of the Responsiveness

treatment and the recruitment conditions, but also whether social motivation or government

responsiveness can drive longer­term engagement in the collaborative management of public services.

Like previous estimations, we fail to reject the null that any social motivation recruitment condition or

that the announcement about reporting by local leadership significantly increased reporting during the last

two weeks of Phase 2. We do find, however, a strong signal that responsiveness from government to the

citizen reports has a significant and positive effect on reporting. The Responsiveness treatment boosted

reporting by Phase 1 reporters 123% and boosted reporting by Phase 2 reporters 32%. This result suggests

that responsiveness is necessary to sustain participation in collaborative governance, even if it is not a

predictor of initial engagement. Indeed, only the responsiveness treatment has a lasting effect on all

reporters that were part of this study.

24

Table 4. Total number of reports submitted by each reporter during the last two weeks of Phase 2

To aid the interpretation of this key finding, Figure 3 shows the proportion of reporters who

submitted solid waste reports in response to each of the 15 prompts during the Phase 2 reporting period.

As can be seen, the effect of the responsiveness treatment is most pronounced at the end of the reporting

period when pooling all reporters. For reporters recruited during Phase 1, responsiveness to reports was

critical for boosting reporting throughout the reporting period. In Appendix E, we show that

responsiveness boosted engagement both for reporters who were active during both halves of the Phase 1

reporting period and for reporters who became deactivated, indicating that responsiveness can both keep

and bring citizens out of the valley of disengagement. For reporters recruited during Phase 2,

responsiveness to reports only boosted reporting for the second half of the reporting period, which is

consistent with the theoretical dynamics highlighted in the conceptual model above.

25

Figure 3. Proportion of reporters responding to each prompt during Phase 2 broken out by phase of recruitment. Legend: red is reporters assigned to the responsiveness condition, grey is reporters assigned to the control condition for responsiveness.

Attitudinal and Behavioral Outcomes on Trust in Government

To test the proposition that responsiveness from government increases reporter beliefs that

government is responsive to their concerns, we fielded a post­reporting survey to measure reporters’ trust

in government and their behavioral willingness to help the KCCA manage services apart from solid

waste. This survey, administered five weeks after the Phase 2 reporting period, intended to measure

behavioral spillover from experiencing responsiveness from government in other areas and to assess

whether the treatment changed attitudes more directly. While responsiveness did strongly influence

week­to­week reporting, it appears from the data that this effect quickly wears off and does not have

26

long­term implications for attitudes about government and willingness to volunteer time to help

government test and create processes for collaborative governance (Figure 4). In no case did volunteers

randomly assigned to the responsiveness condition hold more favorable attitudes about public services or

government, measured by stated satisfaction with solid waste services, perception of the responsiveness of

the KCCA to citizen concerns, and trust in government. Likewise, when reporters were asked to volunteer

their time to help the KCCA develop a more general reporting platform for collaborative governance

across a range of public services, reporters who were subjected to the responsiveness treatment were no

more likely to volunteer either before or after a reminder.

Figure 4. Attitudinal and behavioral responses to the responsiveness treatment five weeks after the conclusion of the Phase II reporting period. Error bars show bootstrapped 95% confidence intervals.

These results demonstrate the importance of continuously reinforcing positive beliefs about

responsiveness at part of collaborative governance arrangements, since deeper attitudes related to trust in

27

government and willingness to assist government in the delivery of services are either difficult to move or

the effects are short­lived. Since we fielded this survey five weeks after the reporting period, we cannot

parse these two possibilities, but the implication of either possibility is the same in light of the large

treatment effect of responsiveness during the reporting period: citizen beliefs about responsiveness need

to be continually reinforced in order to avoid having them fall into the valley of disengagement.

Governments must carefully consider how to reinforce these beliefs if they want to gain the information

possessed by citizens.

Discussion and Conclusions

To improve the provision of public goods and services, governments around the world are

developing new technologies and strategies to collect information from citizens on service delivery and

local conditions — data vital in determining how to allocate government resources and target

improvements where they are most needed and valuable to the public. This information, though difficult

and expensive for government to collect on its own, is readily accessible to citizens. Yet despite

significant government effort and investment to spur collaborative governance with citizens to improve

public services, many of these efforts fail to activate and sustain significant engagement ( Evans &

Campos 2013 , Grossman et al. 2015 ; Grossman et al. 2016 ). We theorize that to foster and sustain citizen

engagement to report on public services, a government needs to both recruit citizens that care more about

the collective goods of the community, and demonstrate its commitment to responding to citizen reporting

efforts. Without the involvement of citizens motivated to improve public goods, and input on local

conditions and where services are most needed, governments struggle to target service improvements in

ways most valuable to citizens. This reduces citizens’ motivations to collaborate with government and

leaves government with limited opportunities to demonstrate its motivations and efforts to improve

services for citizens. As a result, citizens fall into a valley of disengagement, a negative cycle where

citizen trust in government wanes and inactivity is further reinforced. We explore how to break out of this

logjam through nominating citizen reporters that care more about improving public goods, heightening

reporters’ value of improving collective goods through community announcements, and increasing the

responsiveness of government to citizen reporting.

In partnership with the Kampala Capital City Authority in Uganda we investigated how to

activate and sustain the engagement of citizens in collaborative governance. We created a deep and

on­going partnership with a major municipal government, and modified an SMS platform to prompt and

process thousands of spatially­explicit citizen reports about solid waste services. In this study, we find that

citizen monitors treated with social motivation during and after recruitment — a nomination by a

28

neighbor or local authority figure and/or a community announcement with reporters’ names and

responsibilities — did not engage more frequently at any point in the reporting period. In contrast, we

find strong evidence that reporters who experience a responsive government — weekly personalized

messages sharing real government plans to respond to concerns raised by citizens — are significantly

more likely to engage over a period of months.

We initiated this study with the expectation that nomination and community announcements

would enhance the quantity and quality of citizen reporting. Specifically, we expected that nominees

would feel a greater sense of obligation to their community to provide frequent and actionable reports and

may have gained social status by representing the community. Additionally, we expected that nominated

reporters would be better connected and more likely to have leadership attributes, which might also

enhance reporting. In this study, we find no evidence that supports these expectations. From a policy

perspective, these results indicate that governments that are investing in costly screening methods to

recruit citizen monitors will not derive a great deal of value from these efforts. From a more optimistic

standpoint, governments are able to recruit citizen reporters without costly screening and receive similar

reporting results as more intensive recruitment efforts.

From a research standpoint, our findings are some of the first on the limited effectiveness of

tapping into community networks to drive more engagement in collaborative governance. These results

are inconsistent with other findings on the significance of social networks in Uganda, which suggest that

by grouping citizens according to the issues they feel are most important and offering opportunities to

develop collective experiences can significantly elevate participation in SMS­based platforms that

promote discussion of public issues ( Blaschke et al. 2013 ). More broadly, our results offer some caution

about the promise of initiating collaborative forms of governance based on the resources of pre­existing

social networks ( Olsson et al. 2006 ; Tkacheva and Bauhoff 2015 ; Avdeenko and Gilligan 2015 ) and

suggest that future research must address ways to harness community networks to initiate and sustain

collaborative governance, especially where trust in government is low.

In this experiment, we also expected to find that subjects who experienced a government that was

responsive to their reports would exhibit greater and sustained engagement in collaborative governance.

More specifically, citizen monitors who heard about what the government is doing with the data that they

report should feel that their reporting efforts would more likely improve public services in their

communities, and therefore they would respond to more prompts. This study produced strong evidence

that government responsiveness is necessary to sustain reporting on public services over longer time

periods and has wide­reaching implications on policy and research. In places where citizens’ trust in

government to respond is low, government’s efforts to inform citizens on how their reports are being used

29

to improve services should be a critical component any collaborative governance effort. A lack of

responsiveness to citizens’ reporting efforts might explain the relatively low rates of participation found in

e­governance platforms. However, we also learned that any trust or beliefs about the responsiveness of

governments that citizens gained through the responsiveness condition was ephemeral. Other research

suggests that attitudes about government change slowly, and that important factors influencing citizens’

trust in government are perceptions of efficacy in shaping the actions of government ( Parent et al. 2005 ),

government responsiveness to citizens ( Tolbert et al. 2006 , Welch et al. 2005 ), and political­cultural

variables like satisfaction with democracy ( Christensen et al. 2014 ). Our findings indicate that changing

citizens’ trust in government and willingness to participate in collaborative governance likely require a

long­term commitment to government responsiveness and tangible improvements in public services.

Our study also puts the potential for ICT to enhance governance, trust in government, and social

monitoring into context. The potential for ICT to shift the ways that citizens and government interact, to

generate more and better information needed for governance, and to promote a more engaged citizenry in

public life is regularly touted. Research has generated less sanguine expectations about collaborative

governance and social accountability, with recent reviews highlighting the need to better understand

feedback loops and the strategic nature of public engagement in the governance ( Fox 2015 ). We show that

responsive governments cannot depend on ICT to change their relationship with citizens and enhance

collaborative governance without credibly signaling their commitment to act on the information provided

by citizens. By doing so, however, governments are able to break out of the valley of disengagement,

where low citizen engagement and the substandard provision of public goods are reinforcing.

30

References

Ansell, Chris, and Alison Gash. 2008. Collaborative Governance in Theory and Practice. Journal of

Public Administration Research and Theory 18(4): 543­571.

Adserà, Alícia, Carles Boix, and Mark Payne. 2013. Are You Being Served? Political Accountability and

Quality of Government. The Journal of Law, Economics & Organization 19(2): 445­490.

Avdeenko, Alexandra, and Michael J. Gilligan. 2015. International interventions to build social capital:

evidence from a field experiment in Sudan. American Political Science Review 109:3 (August):

427­449.

Banerjee, Abhijit, Angus Deaton, and Esther Duflo. 2004. Health, Health Care, and Economic

Development: Wealth, Health, and Health Services in Rural Rajasthan. The American economic

review 94:2 (May): 326­330.

Banerjee, Abhijit, and Esther Duflo. 2006. Addressing Absence. The Journal of Economic Perspectives,

20(1): 117­132.

Beaman, Lori, and Jeremy Magruder. 2012. Who gets the job referral? Evidence from a social networks

experiment. The American Economic Review 102:7 (December): 3574­3593.

Berardo, Ramiro, and John T. Scholz. 2010. Self­Organizing Policy Networks: Risk, Partner Selection,

and Cooperation in Estuaries. American Journal of Political Science 54:3 (July): 632­649.

Berkes, Fikret. 2009. Evolution of co­management: Role of knowledge generation, bridging organizations

and social learning. Journal of Environmental Management 90:5 (April): 1692­1702.

Besley, Timothy, and Robin Burgess. 2002. The Political Economy of Government Responsiveness:

Theory and Evidence from India. The Quarterly Journal of Economics, 117:4 (November):

1415­1451.

Bhuiyan, Shahjahan H. 2010. A Crisis in Governance: Urban Solid Waste Management in Bangladesh.

Habitat International, 34:1 (January): 125­133.

Björkman, Martina, and Jakob Svensson. 2009. Power to the People: Evidence from a Randomized Field

Experiment on Community­Based Monitoring in Uganda. Quarterly Journal of Economics,

124(2): 735­769.

Blaschke, Sean M., Peter P. Carroll, Daniela Rojas Chaves, Michael G. Findley, Madeleine C. Gleave,

Robert N. Morello, and Daniel L. Nielson. 2013. Extrinsic, Intrinsic, and Social Incentives for

Crowdsourcing Development Information in Uganda: A Field Experiment. Working Paper.

Prepared for submission to the 6th Annual Conference on the Political Economy of International

Organizations, Universities of Mannheim and Heidelberg, Germany.

Brabham, Daren C. 2009. Crowdsourcing the Public Participation Process for Planning Projects. Planning

31

Theory 8:3 (August): 242­262.

Brabham, Daren C. 2012. Motivations for participation in a crowdsourcing application to improve public

engagement in transit planning. Journal of Applied Communication Research 40(3), 307­328.

Brogden, Mike, and Nijhar Preeti. 2005. Community Policing. National and international models and

approaches. New York, NY: Routledge.

Budhathoki, Nama R., and Caroline Haythornthwaite. 2013. Motivation for Open Collaboration Crowd

and Community Models and the Case of OpenStreetMap. American Behavioral Scientist 57:5

(May): 548­575.

Caseley, Jonathan. 2006. Multiple accountability relationships and improved service delivery performance

in Hyderabad City, Southern India. International Review of Administrative Sciences 72:4

(December): 531­546.

Chandler, Dana, and Adam Kapelner. 2013. Breaking monotony with meaning: Motivation in

crowdsourcing markets. Journal of Economic Behavior & Organization 90 (June): 123­133.

Christensen, Tom, and Per Lægreid. 2014. Trust in Government: The Relative Importance of Service

Satisfaction, Political Factors and Demography. Public Performance & Management Review

28(4): 478­511.

Crooks, Andrew T., and Sarah Wise. 2013. GIS and agent­based models for humanitarian assistance.

Computers, Environment and Urban Systems 41(September): 100­111.

Dahlander, Linus, and Henning Piezunka. Open to suggestions: How organizations elicit suggestions

through proactive and reactive attention. Research Policy 43:5 (June): 812­827.

Duflo, Esther, Pascaline Dupas, and Michael Kremer. 2015. School Governance, teacher Incentives, and pupil­teacher ratios: Experimental evidence from Kenyan primary Schools. Journal of Public Economics 123 (March): 92­110.

Evans, Angela M., and Adriana Campos. 2013. Open Government Initiatives: Challenges of Citizen Participation. Journal of Policy Analysis and Management 32:1 (Winter): 172­185.

Fafchamps, Marcel, and Alexander Moradi. 2015. Referral and Job Performance: Evidence from the

Ghana Colonial Army. Economic Development and Cultural Change 63:4 (July): 715­751.

Ferraz, Claudio, and Frederico Finan. 2011. Electoral Accountability and Corruption: Evidence from and

Audits of Local Governments. The American Economic Review 101:4 (June): 1274­1311.

Fox, Jonathan A. 2015. Social Accountability: What Does the Evidence Really Say? World Development

72 (August): 346­361.

Grossman, Guy, Macartan Humphreys, and Gabriella Sacramone­Lutz. 2014. “I wld like u WMP to extent

electricity 2 our village”: On Information Technology and Interest Articulation. American

Political Science Review , 108:3 (August): 688­705.

32

Grossman, G., Macartan Humphreys, and Gabriella Sacramone­Lutz. 2015. Information Technology and

Political Engagement: Mixed Evidence from Uganda. Working paper. Draft Version 1.1.: October

6, 2015. http://www.columbia.edu/~mh2245/papers1/GHS_Scale.pdf .

Grossman, Guy, Kristin Michelitch, Marta Santamaria. 2016. Texting Complaints to Politicians: Name

Personalization and Politicians’ Encouragement in Citizen Mobilization. Comparative Political

Studies (September)

Hawkins, Darren G., David A. Lake, Daniel L. Nielson, Michael J. Tierney. 2006. Delegation and Agency

in International Organizations . New York: Cambridge University Press.

Joshi, Anuradha. 2013. Do they work? Assessing the Impact of Transparency and Accountability

Initiatives in Service Delivery. Development Policy Review 31(s1): 29­48.

Katusiimeh, Mesharch W., Arthur P. J. Mol, and Kees Burger. 2012. The Operations and Effectiveness of

Public and Private Provision of Solid Waste Collection Services in Kampala. Habitat

International 36:2 (April): 247­252.

Kiewiet, D. Roderick, and Mathew D. McCubbins. 1991. The Logic of Delegation . University of Chicago

Press.

Kinobe, Joel R., Charles B. Niwagaba, Girma Gebresenbet, Allan J. Komakech, and Bj ö rn Vinnerås.

2015. Mapping out the solid waste generation and collection models: The case of Kampala City.

Journal of the Air & Waste Management Association 65(2), 197­205.

Leifeld, Philip, Volker Schneider. 2012. Information Exchange in Policy Networks. American Journal of

Political Science 56:3 (July): 731­744.

Linders, Dennis. 2012. From e­government to we­government: Defining a typology for citizen

coproduction in the age of social media. Government Information Quarterly 29:4 (October),

446­454.

Linders, Dennis. 2013. Towards open development: Leveraging open data to improve the planning and

coordination of international aid. Government Information Quarterly 30:4 (October), 426­434.

Mansuri, Ghazala, and Vijayendra Rao. 2013. Can participation be induced? Some evidence from

developing countries. Critical Review of International Social and Political Philosophy 16(2):

284­304 .

McCubbins, Mathew D., and Schwartz, Thomas. 1984. Congressional Oversight Overlooked: Police

Patrols versus Fire Alarms. American Journal of Political Science 28:1 (February): 165­179.

McGee, Rosemary, and Ruth Carlitz. 2013. Learning Study On “The Users” In Technology for

Transparency And Accountability Initiatives: Assumptions and Realities. Knowledge Programme .

http://www.ruthcarlitz.com/pdf/IDS­UserLearningStudyonT4T&AIs.pdf

33

McGuire, M. 2006. Collaborative Public Management: Assessing What We Know and How We Know It.

Public Administration Review 66:s1 (December): 33­43.

Nielson, Daniel L., and Michael J. Tierney . Delegation to International Organizations: Agency Theory

and World Bank Environmental Reform, International Organization 57:2 (April): 241­276.

Oates, Briony J. 2003. The potential contribution of ICTs to the political process. Electronic Journal of

e­Government 1(1): 31–39.

Okot­Okumu, James, and Richard Nyenje. 2011. Municipal Solid Waste Management Under

Decentralization in Uganda. Habitat International 35:4 (October): 537­543.

Olken, Benjamin. 2007. Monitoring Corruption: Evidence from a Field Experiment in Indonesia. Journal

of Political Economy 115:2: 200­249

Olsson, Per, Lance Gunderson, Stephen Carpenter, Paul Ryan, Louis Lebel, Carl Folke, and C.S. Holling.

2006. Shooting the Rapids: Navigating Transitions to Adaptive Governance of Social­Ecological

Systems. Ecology and Society 11:1 (June).

Oteng­Ababio, Martin. 2010. Private Sector Involvement in Solid Waste Management in the Greater

Accra Metropolitan Area in Ghana. Waste Management & Research 28:4 (April): 322­29.

Parent, Michael, Christine A. Vandebeek, and Andrew C. Gemino. 2005. Building Citizen Trust Through

E­government. Government Information Quarterly 22 (4): 720­736.

Powell, G. Bingham, Jr. 2004. Political Representation In Comparative Politics. Annual Review of

Political Science 7 (June): 273­296.

Przeworski, Adam, Bernard Manin, and Susan C. Stokes. 1999. Democracy, Accountability, and

Representation . Vol. 2. New York: Cambridge University Press.

Rotberg, Robert I., and Jenny C. Aker. 2013. Mobile Phones: Uplifting Weak and Failed States. The

Washington Quarterly 36(1): 111­12 5.

Sandstr ö m, Annica, Beatrice Crona, and Ö rjan Bodin. 2014. Legitimacy in Co­Management: The Impact

of Preexisting Structures, Social Networks and Governance Strategies. Environmental Policy and

Governance 24 :1 (January/February): 60­76.

Schultz, P. Wesley, Jessica M. Nolan, Robert B. Cialdini, Noah J. Goldstein, and Vladas Griskevicius.

2007. The Constructive, Destructive, and Reconstructive Power of Social Norms. Psychological

science 18:5 (May): 429­434.

Seidel, Claudius E., Basanta E. Thapa, Ralf Plattfaut, and Bj ö rn Niehaves. 2013. Selective crowdsourcing

for open process innovation in the public sector: are expert citizens really willing to participate?.

In Proceedings of the 7th International Conference on Theory and Practice of Electronic

Governance (pp. 64­72).

34

Smith, Matthew L., Katherine M. Reilly, Yochai Benkler. 2013. Open Development: Networked

Innovations in International Development . Cambridge: The MIT Press.

Speer, Johanna. 2012. Participatory Governance Reform: A Good Strategy for Increasing Government

Responsiveness and Improving Public Services? World Development 40:12 (December):

2379­2398.

Stevenson, William B., and Danna Greenberg. 2000. Agency and Social Networks: Strategies of Action in

a Social Struction of Position, Opposition, and Opportunity. Administrative Science Quarterly

45:4 (December): 651­678.

Tkacheva, Olesya, and Sebastian Bauhoff. 2015. Social Capital and Community Monitoring of Healthcare

Services in Tajikistan. RAND Working Paper. Prepared for the presentation at the Annual

Meeting of the Midwestern Political Science Association, Chicago, IL, April 16­19, 2015.

http://www.rand.org/content/dam/rand/pubs/working_papers/WR1000/WR1097/RAND_WR109

7.pdf

Tolbert, Caroline J., Karen Mossberger. 2006. The Effects of E­Government on Trust and Confidence in

Government. Public Administration Review 66:3 (May/June): 354­369.

Vermeiren, Karolien, Anton Van Rompaey, Maarten Loopmans, Eria Serwajja, and Paul Mukwaya. 2012.

Urban Growth of Kampala, Uganda: Pattern Analysis and Scenario Development. Landscape and

Urban Planning 106:2 (May): 199­206.

Vráblíková, Kateřina. 2013. How Context Matters? Mobilization, Political Opportunity Structures, and

Nonelectoral Political Participation in Old and New Democracies. Comparative Political Studies

(June)

Welch, Eric W., Charles C. Hinnant, and M. Jae Moon. 2005. Linking Citizen Satisfaction with

E­Government and Trust in Government. Journal of Public Administration Research and Theory

15:3 (July): 371­391.

World Bank. 2004. Making Services Work For Poor People. World Development Report. Washington,

D.C.: The World Bank.

https://openknowledge.worldbank.org/bitstream/handle/10986/5986/WDR%202004%20­%20Eng

lish.pdf?sequence=1

35

Supplementary Information

Table of Contents

A. Pre­Experimental Scoping Survey

B. Reporter Recruitment Protocols

C. Prompts Sent to Citizen Reporters

D. Pre­Registered Hypotheses

E. Heterogeneous Treatment Effects of Responsiveness for Inactive and Active Phase 1 Reporters

F. Phase 2 Results by Zone

G. Complier Average Causal Effects for LC1 Announcement treatment in Phase 2

H. Examining the possibility of a social norm treatment effect in the Responsiveness treatment

36

A. Pre­Experimental Scoping Survey

During the summers of 2013 and 2014, we completed preliminary research for the project and

established the partnerships that will facilitate the proposed research. To scope out the relevance of our

project to field conditions in Kampala, we embedded questions about satisfaction with solid waste

services into a broader household survey. In total, we received responses from 439 individuals identified

through a random walk pattern in randomly selected neighborhoods across Kampala. Initial survey data

indicates that Kampala citizens are highly concerned about solid waste services in their communities. For

brevity, we provide summary statistics about three questions: (1) personal concern about waste disposal;

(2) dissatisfaction with current collection services; and (3) self­reports of burning waste at least one time

per week. The vast majority of respondents are personally concerned with the state of solid waste

collection and a majority are actively dissatisfied with the current state of solid waste services in their

neighborhoods. Additional survey responses reveal that a minority of households are able to take

advantage of formal waste collection services, and most households are forced to burn their trash on a

weekly basis. Our survey data suggest that 86 percent of Kampala residents own mobile phones, so

recruitment for monitoring can occur from the vast majority of residents in all zones of the city (Figure

A1).

Figure A1. Resident perceptions of solid waste services and conditions in preliminary survey.

37

B. Reporter Recruitment Protocols

Below are instructions used by the enumeration team to recruit subjects for the study. Recruitment drives

were carried out from November 5­26, 2015 for Phase 1, and June 9­16, 2016 for Phase 2. The

recruitment team comprised approximately 20 Ugandans.

Setting up in the zones (Phase 1 and 2)

When your team first enters a village, inform the village chairperson of the project and secure their

support for any project­related activities, such as an LC1 recruitment or LC1 announcement treatment.

Use the information flyer (see Reporter Recruitment Flier below) and introduction letter to help gain the

support of the LC1. If the LC1 is unavailable, ask him to delegate the responsibility to another local

authority figure, such as the Vice­Chairperson, Secretary of Defence, or member of the Local Council

Committee.

Next, have the chairperson or a resident of the zone describe the boundaries of the zone. Discuss how to

divide up the zone into four cells of roughly similar size, and begin recruitment activities according to the

treatment assigned to that zone. Five individuals will be recruited from each of the four cells. In this way,

20 reporters will be recruited in each zone. [In Phase 1, our recruitment team was asked to recruit three

individuals from each of the four cells per zone, for a total of 12 reporters per zone]. Eligible subjects

must be an adult (over the age of 18), a resident of the zone and the primary user of their own cell phone.

Random Citizen Recruitment (Control treatment for Phases 1 and 2)

For zones that are assigned for random recruitment, the enumeration team will follow a random walk

pattern (see the generating a random walk pattern section below) to select subjects. First, find your way

to the center of one of the four cells in a zone, then follow a random walk pattern for three minutes using

a timer on a phone or tablet. Once the timer reaches three minutes, attempt to recruit the nearest adult. If

the adult is ineligible or refuses to participate, restart the timer and follow a random walk pattern again for

three minutes to select the next potential subject. The same process will be followed until 5 subjects have

been selected in each cell. The work is complete when a total of 20 subjects have been recruited in the

zone. [In Phase 1, our recruitment team was asked to recruit three individuals from each of the four cells

per zone, for a total of 12 reporters per zone]. Use the following steps to sign up the subjects.

1) Introduce yourself and inform the subject about the citizen monitoring program.

38

2) If the subject is interested in participating, read the flyer (see the Reporter Recruitment Flier below)

to the subject in his/her preferred language.

3) Enumerators should not place any pressure on the respondents to participate, including informally

with body language.

4) The subject is under no obligation to respond and may terminate the interview at any time without

consequence.

5) If the subject agrees to participate, survey the subject using the Kobocollect survey.

Neighbor Nomination (Phase 1 treatment)

Contact the first adult in sight. To be eligible, the person must be an adult resident of the zone. Explain the

program, hand them an information slip and answer any questions they raise. Ask them if they'd be

willing to nominate a “reliable and trustworthy” person from the zone to become a reporter in the system.

Follow the steps to sign up a subject in the previous section. If so, ask the person to make a personal

introduction to the nominee either by calling the person or by making a face­to­face introduction. Make

sure the nominated individual is an adult resident of the zone. Explain the program to the nominated

individual, hand them an information slip and answer any questions they raise. Ask them if they'd be

willing to participate as a reporter and remind them they have been nominated by a neighbor. If yes, sign

them up using the survey on KoboConnect. Ask the person if they would like to nominate anyone to be a

reporter, regardless of whether they have signed up or not. If no, again randomly walk for 2­3 minutes.

Repeat the sign­up process.

LC1 Announcement (Phase 2 treatment)

Recruit subjects using the recruitment method assigned to the zone (see Random Citizen Recruitment or

LC1 Recruitment). Additionally, inform the subject that in an upcoming zone meeting, the LC1 will

announce them as a citizen monitor selected to represent the zone. After all 20 reporters have been

recruited in a zone, provide the LC1 with a list of the names of those selected to be citizen monitors.

Secure the LC1s commitment to announcing the program and names of citizen monitors at an upcoming

zone meeting. Lastly, complete the LC1 Announcement survey on Kobocollect to gather the LC1s contact

information. The implementation team will contact LC1s by phone one week following the completion of

the recruitment activity to remind the LC1s to make the announcement at a zone­wide meeting.

LC1 Recruitment (Phase 2 treatment)

Subjects in zones assigned to Treatment 3 will be recruited by the LC1. The LC1 will personally

39

introduce the subject to the recruitment team and recommend them as a citizen monitor. Once the

recruitment team has been introduced to the subject, follow the instructions below to sign up a subject.

Reporter Recruitment Flier (Phase 1 and 2)

Invitation to Report on Solid Waste Collection in Your Neighborhood!

We are an independent research group launching a project that will allow residents of Kampala to use

SMS to report on waste management issues in their neighborhoods. Your input is very valuable and we

hope you will participate in making Kampala a cleaner and more livable city. We are asking you to join

the platform.

If you sign up to be a reporter, we will send you 2­3 messages per week over 8 weeks asking you to report

on the solid waste condition and services in your neighborhood. Each week there will be a lottery to win

airtime.

All messages that you send and receive from us will be toll­free and will not reduce your airtime. If you

ever have questions, you can send the message “HELP” to 6585. Someone will contact you to answer

your questions. You can also send the message “STOP” to 6585 at any time to stop receiving messages.

Your name or contact information will not be shared with anyone. Your responses will be used to inform

the Kampala Capital City Authority about which areas of Kampala require better waste management

services. Please contact Jacob Skaggs (0780291311) if you have any questions or concerns about the

program.

Generating a random walk pattern

1) Find an intersection in each of the assigned cells. An intersection is the crossing of any road, path, or

alley that leads to the entrance of residential dwellings. The starting intersection should be located by

walking several minutes into the assigned cell.

2) Assign each direction leading from the intersection a number. Roll the dice and move in the direction

selected randomly.

3) Any time you reach another intersection, assign each direction that moves forward from your walk

path a number and roll the dice, moving in the direction selected randomly. You should only turn around

if you reach a dead end or the edge of the assigned cell.

4) The only reason that the randomly chosen direction should not be an option is if you have already

been down a path and you know that it leads to a dead end.

40

C. Prompts Sent to Citizen Reporters

Prompts from Experiment 1:

1. How many times have you observed waste being picked up and removed from your zone in the last weeks? [REPLY with a number]

2. How many waste heaps have you observed being burned in your zone during the last 24 hours? [REPLY with a number]

3. Please describe the location of any waste heap that needs attention from the KCCA or its contractors. [REPLY with a location description]

(In the Experiment 1, each of the three messages above were sent to all subjects once each week over a period of 8 weeks).

Prompts from Experiment 2:

1. Does a rubbish truck come to your neighborhood? 1) no 2) yes 3) don’t know 2. When did the rubbish truck last collect your rubbish? A) never B) more than two weeks ago C)

last week D) this week 3. What is the most common way for your neighbors to dispose of their rubbish? 1) burn rubbish 2)

throw in a rubbish pile 3) throw in a ditch 4) use a rubbish truck 5) don’t know 4. How happy are you with rubbish collection services? 1) very unhappy 2) unhappy 3) neither

happy nor unhappy 4) happy 5) very happy 6) don’t know 5. How often do you see rubbish spilling from rubbish trucks? 1) never 2) rarely 3) two times a

month 4) once a week 5) many times a week 6) don’t know 6. How much waste is there on the ground in your neighborhood? (1) none (2) some small piles (3)

a few larger piles (4) waste in many places 5) don’t know 7. On the path you walk in and out of your zone, how many waste piles would you see? [Respond

with a number] 8. In a typical week, how many times would you see burning rubbish if you walked in the zone for

fifteen minutes per day? 9. How often does the rubbish truck collect rubbish on the chosen day of the week? 1) never 2) not

often 3) often 4) very often 5) don’t know 10. How happy are you with how often your rubbish is collected? 1) very unhappy 2) unhappy 3)

neither happy or unhappy 4) happy 5) very happy 6) don’t know 11. How happy are you with the distance from your home to the rubbish truck? 1) very unhappy 2)

unhappy 3) neither happy or unhappy 4) happy 5) very happy 12. How well do rubbish collectors treat you? 1) very bad 2) bad 3) neither bad nor good 4) good 5)

very good 13. What is the biggest problem with your rubbish collection service? [open response] 14. Are there any other rubbish or sanitation services that you would like? [open response] 15. Please describe how to reach the largest rubbish pile near your home. [open response]

41

D. Pre­Registered Hypotheses

Experiment 1

H1a: More nominated reporters will respond to at least one prompt than randomly recruited reporters. H1b: Nominated reporters will respond to more prompts than randomly recruited reporters, measured as a count both over the entire 8­week experiment and within individual weeks. H1c: Nominated reporters will respond to more open­ended prompts than randomly recruited reporters, measured as a count both over the entire 8­week experiment and within individual weeks.

Experiment 2

H1a: More reporters assigned to the LC1 recruitment treatment will respond to at least one prompt than randomly recruited reporters. H1b: More reporters assigned to the LC1 announcement treatment will respond to at least one prompt than reporters in the announcement control condition. H1c: More reporters assigned to the responsiveness treatment will respond to at least one prompt than reporters in the responsiveness control condition. H2a: Reporters assigned to the LC1 recruitment treatment will respond to more prompts than randomly recruited reporters, measured as a count both over the entire 8­week experiment and within individual weeks. H2b: Reporters assigned to the LC1 announcement treatment will respond to more prompts than reporters in the announcement control condition, measured as a count both over the entire 8­week experiment and within individual weeks. H2c: Reporters assigned to the responsiveness treatment will respond to more prompts than reporters in the responsiveness control condition, measured as a count both over the entire 8­week experiment and within individual weeks. H3a: Reporters assigned to the LC1 recruitment treatment will respond to more open­ended prompts than randomly recruited reporters, measured as a count both over the entire 8­week experiment, within individual weeks (to measure changes in participation over time), and for the final two weeks (to measure attrition). H3b: Reporters assigned to the LC1 announcement treatment will respond to more open­ended prompts than reporters in the announcement control condition, measured as a count both over the

42

entire 8­week experiment, within individual weeks (to measure changes in participation over time), and for the final two weeks (to measure attrition). H3c: Reporters assigned to the responsiveness treatment will respond to more open­ended prompts than reporters in the responsiveness control condition, measured as a count both over the entire 8­week experiment, within individual weeks (to measure changes in participation over time), and for the final two weeks (to measure attrition).

43

E. Heterogeneous Treatment Effects of Responsiveness for Inactive and Active Phase 1 Reporters

As an extension to our main analysis, we are interested in whether responsiveness can both cause

more persistent reporting among activated reporters, as well as activate or re­activate reporters who fell

below the engagement threshold that we theorize above. Thus, we divide all Phase 1 reporters who were

prompted during Phase 2 for reports into three subgroups: (1) reporters who were active in both the first

and second half of the reporting period during Phase 1 (the “activated” sub­group); (2) reporters who

were only active in the first half of the reporting period during Phase 1 (the “deactivated” sub­group); and

(3) reporters who were never active during Phase 1 (the “inactive” sub­group). Table F1 displays

reporter­wise regression results for the number of reports received in total and during the last two weeks

of Phase 2, where Inactive reporters and the control group are the baseline condition.

Table E1. Reporting by Phase 1 reporters during Phase 2 by activation status.

The results show that responsiveness does not boost rates of reporting for inactive reporters. For

reporters that were inactive for the entirety of Phase 1, the responsiveness condition has no effect on

44

reporting. In contrast, responsiveness further boosts reporting for both activated and deactivated Phase 1

reporters, indicating not only that responsiveness can keep reporters out of the valley of disengagement,

but also that it can reactivate those who have fallen below the activation threshold for collaborative

governance.

45

F. Phase 2 Results by Zone

Here we report results both for the pooled group of subjects recruited during both Phase 1 and 2,

as well as the results split by the recruitment phase at the zone level. We found that six of the seven zones

in Phase 2 that our enumerators recruited 15 or fewer reporters were subsequently assigned to the

Responsiveness treatment. We thus also examine the subset of zones with 16 or more reporters for the

split Phase 2 analysis (recall the target was to recruit 20 reporters per zone) where the number of reporters

at the zone­level is balanced by treatment condition. In all tables reported below, the base conditions are

Random Citizen recruitment, the control condition for the LC1 announcement about reporters, and the

control condition for the Responsiveness treatment.

Considering first the total number of reporters during Phase 2 who submitted at least one,

on­topic report about solid waste management during the eight­week period, only the Responsiveness

condition boosts the number of active reporters as hypothesized (Table 2). In both the pooled and split

models, zones assigned to the responsiveness condition have approximately one extra reporter who is

active on average than zones assigned to control (across all zones and experimental conditions, the mean

is approximately five active reporters per zone during Phase 2). This translates to a 20 percent increase in

the mean number of reporters and indicates substantive as well as statistical significance. This result

indicates that hearing about what the government is doing with the reports via targeted outreach can help

initiate engagement in citizen reporting. In contrast, we do not observe any differences in the number of

active reporters when recruiting by either neighbor or LC1 nomination, or when reporters expected the

LC1 to make an announcement about the platform and reporters at a community meeting. Thus, the

evidence suggests that social motivations are not effective at activating reporting on public services in this

context.

46

Table F1. Total number of active reporters by zone during Phase 2

Turning to the total number of reports made by zone during the 8­week Phase 2 reporting period,

we find very similar results, with only the responsiveness treatment driving more reports (Table 3).

Pooling zones across recruiting periods, we find that the Responsiveness treatment increased the number

of reports per zone by approximately 6.6 over eight weeks (across all zones and experimental conditions,

the mean is approximately 32 reports per zone during Phase 2). This result is largely driven by the

significant effect the Responsiveness treatment had on zones where recruitment took place during Phase 1,

where the treatment increased the number of total reports by zone by approximately nine (Model P1). In

contrast, the Responsiveness treatment did not increase the total number of reports among Phase 2 zones

in ways that are highly inconsistent with random chance (for P2 and P2 CS models, p=0.10~0.15). Like

the results for the total number of active reporters, we do not observe any differences in the number of

total reports by zone when recruiting was done by either neighbor or LC1 nomination, or when reporters

expected the LC1 to make an announcement about the platform and reporters at a community meeting.

47

Table F2. Total number of reports by zone during Phase 2

Finally, we consider the total number of reports per zone by treatment condition for the last two

weeks of the 8­week reporting period. As pre­registered, we are interested not only in the total effects of

the Responsiveness treatment and the experimental recruitment conditions, but also whether social

motivation or government responsiveness can drive longer­term engagement in the collaborative

management of public services. Like previous estimations, we do not find that any recruitment condition

or that the announcement about reporting by local leadership significantly increased reporting during the

last two weeks of Phase 2. We do find, however, a strong signal that responsiveness from government to

the citizen reports has a significant and positive effect on reporting across all zones in both of the field

experiments, actually having a substantively similar effect. This result suggests that responsiveness is

necessary to sustain reporting and participation in collaborative governance, even if it is not a predictor of

initial engagement. Indeed, only the responsiveness treatment has a lasting effect for the entirety of the

reporting period for reporters recruited during Phase 1 and for the end of the reporting period for reporters

recruited during Phase 2.

48

Table F3. Total number of reports by zone during last two weeks of Phase 2

49

G. Complier Average Causal Effects for LC1 Announcement treatment in Phase 2

The potential reporters in the zones assigned to the LC1 Announcement treatment were fully

informed that their names and responsibilities would be announced at an upcoming community meeting.

We delivered the list of recruited reporters to all zone chairs in this treatment condition and asked them to

make such an announcement. We followed­up with a reminder one week after delivering the list of

reporters. At the end of the reporting period, we made three attempts to call all 50 LC1 chairs who had

been asked to make this announcement at a community meeting. We were able to reach 42 chairs and

learned that 16 had made the community announcement and 26 had not made the community

announcement. Those who did not make the community announcement reported that they were busy,

were away from the zone, or did not remember, among other reasons.

In the main results above, we report intent­to­treat estimates that do not take into account the

actual delivery of the LC1 Announcement treatment. Here we estimate complier average causal effects via

2­stage least squares, where the treatment assignment used as an instrument for the delivery of treatment.

Because we did not deliver the names of reporters to LC1 chairs in zones assigned to control, we rule out

two­sided non­compliance. We were not able to collect information about compliance for 8 of the 50

zones assigned to treatment, so we estimate the bounds of CACE. Table G1 drops the zones with missing

compliance information. Table G2 assumes that all zones with missing compliance data are compliers.

Table G3 assumes all zones with missing compliance data are non­compliers. All tables show the second

stage estimates. In no case do we find treatment effects that diverge in substantive or statistical

significance from the intent­to­treat results reported in the main text.

50

Table G1. CACE for LC1 Announcement Condition with missing compliance data dropped

Table G2. CACE for LC1 Announcement Condition with missing compliance data assumed to be in compliance

51

Table G3. CACE for LC1 Announcement Condition with missing compliance data assumed to be out of compliance

52

H. Examining the possibility of a social norm treatment effect in the Responsiveness treatment

As part of the Responsiveness treatment, we informed all reporters about the total number of

reports received from their zone during the previous week. Although reporters were never informed about

the total number of other reporters in their zone, which makes it difficult for the reporters to interpret the

raw number of reports as a social norm, it is nonetheless possible that this information introduced a

bundled nudge toward a social norm into the treatment. Other research finds that when given information

about social norms, on average people revert to the social norm, whether they are above or below it. For

example, Schultz et al. ( 2007 ) find that provider households with descriptive norms about energy

consumption in their neighborhood decreased usage among high­usage households, but had the opposite

effect among low­usage households. If a social norm were activated in our case, we would expect it to

boost reporting from inactive reporters and decrease reporting from active reporters. Since the

Responsiveness treatment is intended to boost reporting for both types of reporters but would be

counteracted by a social norm among active reporters, we examine whether the Responsiveness treatment

has a null or a negative effects among reporters who were active during the previous week. Intuitively,

this transforms the dependent variable to the rate of continued reporting among reporters who were active

in the immediate past. If there is a positive treatment effect in this group, we add confidence that

responsiveness from government is driving the positive treatment effect that we observe in the main

results.

To complete this robustness check, we examine only the reporting behavior for reporters who

were active during the previous week, who either would not be able to interpret whether they were above

or below the social norm or who knew that they were above it. We first calculate the number of weeks

that are eligible for analysis for each reporter, which is the number of weeks when a reporter was active

during the previous week. We then sum the number of reports that these reporters submitted during these

eligible weeks and model treatment effects as above, controlling for the total number of eligible weeks at

the reporter level. Table H1 shows the total number of eligible responses received during the Phase 2

reporting period as a function of treatment assignments. Table H2 shows the total number of eligible

responses received during the last four weeks of Phase 2 reporting period as a function of treatment

assignments. In all models, the results match the main results that use a larger sample of subjects, which

adds confidence that our treatment effects are coming about because of responsiveness from government.

53

Table H1. Total number of reports during Phase 2 among reporters who were active the previous week

54

Table H2. Total number of reports during the last four weeks of Phase 2 among reporters who were active the previous week

55